The test for epistemic uncertainty

Reference to published code

This is the test for epistemic uncertainty. The data features are coordinates of three randomly selected points and the targets are areas of triangles built on them.

All arguments are defined in interval $[0, 100]$.

KAN makes a model with 4.5% accuracy. When compared to neural networks it is more accurate. The problem lays in features uncorrelated with targets. The points could be sitting near right top area of the definition field and expressed by large numbers, while the area can be relatively small.